The logic is flawed with respect to gres/mps and cloud based nodes.

without this change GRES with large count values would
have those values silently changes (i.e. Count=4G changed
to a count of 0).

If we get an error on one record, try to keep loading others
to at least validity check them

The RPC change only happened in v19.05

without this change if gres/mps is configured in gres.conf without
files then gres.c logic will try to access the topo array, which
is NULL.

this keeps the gres data structures consistent in slurmctld when their
  count changes or is inconsistent with slurmd
also kill jobs allocated to gres with files if their count changes.

If a node's GRES state was recovered, this logic would see
the record and clear it's configured count resulting in
invalid node GRES information.

bug 6304

This is important for gres/mps, where the
bitmap size and GRES count will differ

bug 6304

test33.1 recently broke. likely due to commit 03f8f1d3

Cleanup debug messages while here.

Remove stray #undef HAVE_SYS_SOCKET_H while here, this appears
to date back to before config.h was split off.

The slurm_addr_t typedef comes in from slurm.h already, so this
header is redundant.

NULL check is implied already, and this format is prevalent.

Also remove unnecessary casts while tidying up.

Change to plugrack_new_t to make sure every location is caught.

Remove plugrack_set_major_type().

Was never used to enable the paranoid behavior.

No functional change, aside from switching to __func__ in log messages.

Also switch to xmalloc pattern of:
    int *data = xmalloc(sizeof(*data));
while cleaning these up.

Remove now-unused variables missed in e7a34806/c10f2cf0.